minitok: AI coding with a verification loop, not just code generation

minitok is a repository-aware coding workflow CLI that adds research, planning, implementation, deterministic verification, review, repair, and evidence recording to AI-assisted development. Unlike unstructured code-generation tools, it gives every repository change a repeatable control loop. It supports Anthropic, OpenAI, Google, OpenRouter, and OpenAI-compatible providers, with local-first execution and opt-in telemetry disabled by default. This launch focuses on traceability,

I built minitok because generating a diff is only one part of repository work. The difficult part is understanding the existing codebase, making a plan, running the project’s deterministic checks, reviewing failures, repairing the change, and preserving an inspectable record of what happened. minitok separates those stages into a repeatable workflow. Start with `minitok doctor`, initialize a disposable repository with `minitok migrate`, and preview a task with `minitok run --dry-run`. I’m looking for feedback on the workflow boundaries, verification model, and evidence format. minitok is paid proprietary software; provider costs apply, and it does not guarantee correctness or deployment outcomes.